That's why LLM will eventually be used only for initial interaction between the user in their language, to prepare the data to a specialized model.
Imagine face recognition to work like a text chat, where the PC gets the frame from the camera and writes in the chat: "Who's that? Here's the RGB888 image in hex: ...".
Huh? The images are tokenized in the same way language is and it’s just fed into one single model. Not multiple smaller expert models.
Image gets rasterized into smaller pieces (eg 4x4 pixels) and each of those is assigned a token, similarly how text is broken up into tokens. And the whole thing is fed into a single model.
> Imagine face recognition to work like a text chat, where the PC gets the frame from the camera and writes in the chat: "Who's that? Here's the RGB888 image in hex: ...".
The experts in MoEs aren't specialized in any meaningful task sense. From level of what we would think as tasks MoEs are selected essentially arbitrarily per token and per block.
It’s unsupervised, yes, but “unspecialized in any meaningful task sense” is incorrect, that’s the whole point. It’s just not in the sense of “this is a legal expert, this is a software developer”.
Now do the equivalent of just in time compilation. Claude sees that we need to respond to a lot of pings and writes a program to compute it instead of thinking about each one.
Wouldn't this be faster with an agent skill that has code?
/skill-creator [or /create-skill] Write an agent skill
with code script(s) that use an existing user space IP library that works with your agent runtime, to [...]
You could read about that in 1992 "A Fire Upon the Deep" by Vernor Vinge. There is prompt injection in communication, in the book certain protocols for information communication can not be deterministic so if someone is too smart you get hacked.